TUESDAY, JUNE 19, 2018
Font Size
Topic Shifting mean vs. Reducing variation

Shifting mean vs. Reducing variation

Home Forums Old Forums General Shifting mean vs. Reducing variation

This topic contains 32 replies, has 16 voices, and was last updated by  AB 8 years, 7 months ago.

Viewing 33 posts - 1 through 33 (of 33 total)
  • Author
  • #162800

    I have read many forum entries on both shifting mean and reducing variation but now, I am somewhat confused:
    Is is easier to shift mean or to reduce variation in a process?


    Depends on the process


    Patrick:  While I agree with Stan that which is easier to affect depends upon the process, the more important question should be – which aspect of improvement is more important?  Generally, it will be more important to reduce the variation first, then move the mean (we call this 2 step optimization).  This is generally because the system was not designed to be robust to the sources of variation – all to often those designing the system get it to work at the nominal (mean) level and dump it off on operations.


    Generally, in my experience, moving the mean is much easier than reducing the variation.  If the desired process target is known, than it is often a first order to adjustment to a process input, such as an input temperature or time, or another similar adjustment in order to reset the process to a desired setpoint.
    Reducing variation requires even more in-depth knowledge about the process, in most instances.   In some cases, elimination of out of control points will reduce variation levels to some degree.   If the process is under SPC control, than reductions in process variation often require the process to be changed in a more fundamental way in order to reduce variation.   This typically requires process experimentation in the form of DOE’s that take time and resources, or adding new equipment which has better overall capabilities than what is used now.   Newer equipment often works, but has the disadvantage of added costs that have to be taken into account.


    That is a very well-stated response. And “target” is a key word.
    This question always brings up the image of someone at target
    practice. It is fairly simple to tweak the sight on a rifle (adjusting the
    mean). But variation still depends upon marksmanship.
    That analogy holds true with respect to the points you made.


    StuW, generally I like your answers but your “generally” is way to
    general.It depends on the type of process and the level of process
    knowledge. Some processes are simple settings and, yes, if I can
    just make an adjustment and I know the relationship between the
    adjustment and the desired (and undesired) responses, mean is the
    easiest.That is not the reality in many, many industries. Chemical and food
    processing come to mind, but there are loads of processes out
    there that do not meet the criteria cited above.The right answer is it depends. Depends on knowledge of
    input/output relationships and the ability to tweak the inputs.”Generally” in this case is just as dumb as saying “generally”
    processes shift and drift 1.5 sigma.


    I used “generally” in the earlier post as most of my experience is in the semiconductor industry, and I am giving my insights into the question posed.   From what I have seen, it is ususally much easier to adjust a tool to a target then to figure out ways to reduce variation, which is typically much more challenging to do.  For example, many diffusion furnace processes can use slight time adjustments to reset processes to a target (average oxide thickness achieved).  Variations, however, are more dependent upon temperature control over the entire load and are much more tricky to address. 
    I am interested, though, in hearing about other situations, and types of processes, where reduction in variation is easier to accomplish than setting the process back to a specified target.


    Chemical processes are an excellent example. There are known reactions, that if we set up the conditions right,
    they just happen. But in most facilities they don’t as well as they
    could. It is all about taking the variation out of the critical inputs
    and the reaction will move toward it’s theoretical yield. There is a huge world out there beyond semiconductors, most of
    which doesn’t have .0001 the process knowledge that has to exist
    in a sub-micron world. They also do not have the elaborate
    infrastructure investment that you take for granted. That world is
    all about taking the noise out.


    Reputation - 2
    Rank - Aluminum

    Simple, stop the process and you just shifted the mean. What’s more
    important.. it depends. Are you trying to safe lives, save a customer,
    or just improve or both. Special cause analysis and subsequent change can have a major
    impact to the mean. Common cause can be commonly impacted by
    target proactive audits. So what’s critical, what’s controllable, what needs to be changed in
    the end?


    Our organization was all about taking noise out of processes.  Because the Semi world is so capital intensive, the way most older fabs survive is by taking the variability lower on a tool and finding out how to make it perform even better than the manufacturer proposed in their specs.   There are many 20-30 year old tools still in use where the process was tweaked using DOE’s to allow the results to have reasonable capability even now.   At some point, though, there are limits as to what can be achieved using this approach.  A point where cost cutting and driving improvements on older platforms is no longer the most effective approach.
    The answer to the original question is still the same from my perspective.  It was much easier to reset a process to target than to reduce process variation in most cases.  I understand other situations are possible, and would like to hear more about those.


    Going under the assumption that the process is not capable and thus there is a desire to improve, my first shot would be at the variation. Moving the location of a process with wide variation moves the entire distribution, not a single value. Thus one tail or the other will now be extended beyond where it was prior to the mean shift. I would rather move the mean of a narrow distribution rather than a wide one. And if the reducing of variation deals with a tail, the mean automatically shifts if that is reduced.


    First, are you as good looking as you sound? Secondly, I think that while it is “easier” to shift the mean, it is not necessarily better. Consider the following: Say I am turning three pieces of round stock to a mean diameter of 6″. For each part I get the following diameters: 5.998″, 6.002″, and 6.010″. Everything else being equal, it would be easier for me to make an adjustment on the lathe and change the mean diameter to 4″. However, this might not meet the customer’s requirement (say that requirement was 6″). It would be more diffucult to reduce the variation around the 6″ diameter mean because I may have to purchase a new piece of equipment to be able to do that or I would have to hire a more skilled machinist, etc. Hope that helps.


    Reputation - 0
    Rank - Aluminum

    Horrible example


    And wishful thinking Rachel. What, you like his font? Goodness girl, find some relief!


    Ease has very little to do with anything.  At best it is a secondary consideration.  You need to determine whether a shift in the mean, variation reduction or both will accomplish your objective.  Your strategy should be based on what actually solves the problem.


    “Ease has very little to do with anything.” Interesting statement considering that the original question posted by Patrick was, “Is it easier to shift mean or to reduce variation in a process?”. You should always remember the voice of the customer (Patrick in this case). While this is not a real world question, I happen to know that this was a black belt certification exam question – verbatim. I know because I passed the exam. So, it deserves a direct contextual answer. Never lose sight of the voice of the customer. The customer asked which was easier. Shifting the mean is easier generally (see my example in the previous posting for why).   


    In the transactional world where I live, it’s generally much easier to go after variation than mean. Lots of what I deal with is cycle time reduction against a 1-sides spec limit – for starters, mean is the wrong measure since nearly every distribution I’ve ever seen has been skewed right, so we’d be talking Median reduction – the levers we go after deal more with non-standardized work, non-value added activity, and the special causes that were behind the datapoints out to the right of the distribution. Once you take care of things like that, reduce the variation, the mean/median shift takes care of itself.So Stan is once again correct – it very much depends on the process…


    Reputation - 0
    Rank - Aluminum

    Rachel Honey, Patrick actually stated ” I have read many forum entries on both shifting mean and reducing variation but now, I am somewhat confused”   Then Patrick asked, which was easier, stated nothing about an exam question that you refer to, so highly probable your question as an exam question is purely coincidental. To that point your example was horrible and illustrates a bad process nothing more.  Too many variables exist in your example and in fact it may be easier to change the process. Feed and speed of the cutting system is a very quick easy change that can affect variation, cutting fluids, too much, too little, chemical composition, all easily changed and tested. Tool geometry, tool coatings, etc, and on and on. So to say it is simply easier to change the Mean is a wrong and bad thinking.
    So Stan is correct it depends on the process.


    Siggy,Stan hasn’t been right in 10 years so why should he start now.
    As for your situation.  You are correct in that if you are dealing with cycle time the distribution will be skewed because of the natural limit at 0.  But, I beg to differ.  I can move the center of the distribution very easily by merely throwing bodies at the cycle time.  That is a quick fix for reducing it.  Getting folks to do things in a standardized fashion, the identification and elimination of NVA work and the investigation of special causes are not easy things to change or implement. 
    But, you are correct in going after the variation because its reduction will automatically reduce your cycle time by eliminating the outliers.  That is a longer term and better solution than throwing bodies and resources to reduce the time.


    What’s with all this Stanile Envy????


    Reputation - 0
    Rank - Aluminum

    Gotta give a dog a bone every now and then.


    Ok. First, thanks for the “honey” comment. I appreciate that.
    Let’s speak generally here. There’s no “it depends”. Is it easier or not? Generall, it is easier to shift the mean. That’s why a whole science was developed around six sigma to reduce variation. If it was easier to reduce variation than to shift the mean we would be calling this website “x-bar.com”. But I digress. I come from a manufacturing background so let’s play pretend for a bit. Let’s say I’m shooting a gun at a target – a rifle not a shotgun. I shoot ten shots and end up with a mean distance from the center of the target of 6″. All I have to do to shift the mean is to point the gun a little to the left, say. Now I take ten more shots and my mean distance from the center of the target is 2″. The variation has not changed because I have nearly the same spread in shots. That was pretty “easy”. I’m not saying I’ve improved the process or more fully captured the voice of the customer or anything of that sort. In order for me to reduce the variation in the spread of the shots, I would have to take “hard” steps, like become a more adept marksman, fire the shots in a vacuum to eliminate air density issues, etc, etc. You know what I mean? I mean to say “it depends” is fine if you just want to dodge that bullet but let’s be concrete here.


    Chad makes a valid point…


    Rachel: I agree with you. Shifting the mean is easier. But this is a meaningless question that has no practical siginificance beyond the academic. I like your use of the two examples you gave.


    Reputation - 0
    Rank - Aluminum

    Rachel since your from a Mfg environment then surely you understand that one cannot go around on a daily basis shifting the mean of the process. It is not easier and the risk of sending bad product to the customer is way too high.
    Your example of firing a gun at a target is somewhat misleading as you are assuming that the rifle will fire at the same variation around the target each time without assignable cause. In this example you are correct moving the mean would be an easy fix and anyone who has fired a rifle as much as I have would agree. However in a mfg process variance is rarely predictable, if it is, then there must be an assignable cause, and should always address the variation at this point.


    Its easier to move the mean….variation is alittle harder.  Have to identify the right xs in your process to get that tightened up.


    If I may really digress….you dont just move the gun right or left to adjust center…thats total kentucky windage, Bad solution.  Instead you adjust the gun sights and ensure proper rifle position in the shoulder pocket for centering.  Variation…or “shot group”….now to your point thats the hard Xs to identify….things like breathing, grip tension, air temp, consistent cheek to stock placement. 
    It is easy to improve…..if you have a good shot group but bad centering….it is harder if you have good centering but bad shot group.


    As Chad stated, you can’t rely on mean-shifting to get yourself out of trouble…esp in an mfg environment…The mantra of variation reduction/mitigation/elimination is what’s key here.Variation is the devil and it must be stamped out wherever, whenever possible.wow…


    HeeBee:  I’m normally in-line with your statements, but on this one, I’ve got to disagree.  You may not need to reduce the variation if you have an instance where the Cp is very good (depending on the system could be 1.7 and up) but the Cpk is poor.  In this case, it is a matter of moving the mean.   This is rarely the situation, but could exist.


    Ok, I got a lot of different information (Thanks!) and it seems to me that: shifting the mean, even though it is easier, is not the right thing to do; reducing variation is the right thing to do.  Which is why we are paid the big $$ to do.


    bottom line is, we don’t have that data in evidence here. What I reacted to, was another in a long line of posts and real world situations in which the BB, or Stakeholder wants to simply mean shift PERIOD…tweaking the knobs may result in greater unintended negative consequences.It is a trend that I am addressing.


    The OP is not my customer and I am under no obligation to listen to his voice
    Even if he were my customer, the customer may not always be asking the right questions to truly gain what they are seeking.  So if they ask for either A or B and I truly understand that what they need is C how have I served them by staying only within the choices they have laid out?
    Whether the question showed up on an exam or not is completely irrelevant to this discussion, but thanks for sharing.
    Whether you passed the exam or not is even more irrelevant since it does not automatically prove you answered that question correctly.
    I am aware of the generic answer to the generic question.  It has about as much value as a barrel of dog snot.
    If the first two things that come to your mind when discussing reducing variation on a turned diameter are buy new equipment or hire a more skilled machinist then you haven’t learned a thing despite what your exam may have shown you.


    Hi every one
    I get confused plz tell me if  I am right or not
    I used to know that six sigma project targeting Y this means I  try to change mean  by changing inputs affecting Y  BUT  how I can change variation around means plz tell me

Viewing 33 posts - 1 through 33 (of 33 total)

The forum ‘General’ is closed to new topics and replies.

5S and Lean eBooks

Six Sigma Online Certification: White, Yellow, Green and Black Belt

Six Sigma Statistical and Graphical Analysis with SigmaXL
Six Sigma Online Certification: White, Yellow, Green and Black Belt
Lean and Six Sigma Project Examples
GAGEpack for Quality Assurance
Find the Perfect Six Sigma Job

Login Form